qwen2.5-3b-instruct/train_results.json

{
    "epoch": 3.0,
    "total_flos": 1.0308479847759872e+17,
    "train_loss": 0.4208552807989255,
    "train_runtime": 5691.1072,
    "train_samples_per_second": 9.488,
    "train_steps_per_second": 0.037
}